The potential effects of anxiety on the presentation, course, and treatment response of patients with bipolar disorder have rarely been examined. This is true, although anxiety is a frequent companion of major depression and one that is clearly associated with more severe symptoms of depression and greater functional impairment (1–3), poorer short-term and long-term outcomes (4, 5), a higher rate of help seeking for both psychological and medical problems (6), as well as a greater risk for suicide (7). Furthermore, anxiety has been found to compromise the treatment of major depression. In multiple studies, patients with concomitant anxiety showed a slower or poorer response to antidepressant medication or interpersonal therapy than patients without anxiety symptoms (2, 8–12).
Investigations along different lines of research suggest that anxiety may also be of clinical or theoretical importance in bipolar disorder. Epidemiologic Catchment Area survey data showed individuals with bipolar disorder to be more likely to meet criteria for lifetime panic and obsessive-compulsive disorder (OCD) than individuals with unipolar depression and individuals without bipolar or unipolar depression; the lifetime prevalence of panic disorder and OCD in subjects with bipolar disorder was substantial (13, 14). Young and colleagues (15) found that patients with bipolar disorder and high levels of anxiety symptoms were more likely to exhibit suicidal behavior, alcohol abuse, and cyclothymia and tended to be more likely to require treatment with antidepressant medication or neuroleptics in addition to lithium. Thirty-one percent of patients in this study met diagnostic criteria for panic disorder, generalized anxiety disorder, or both. Finally, evidence from genetic research showed that panic disorder may be a marker for a genetic subtype of bipolar disorder, which prompted the authors to recommend an examination of the differential treatment response of bipolar disorder with and without panic disorder (16, 17).
Although factors related to treatment response in bipolar disorder have been examined in a number of studies, none has included anxiety as a potential correlate of outcome. Given the adverse impact of anxiety on the treatment of unipolar depression and the paucity of data on the potential role of anxiety in bipolar disorder, we sought to determine the effect of anxiety on the acute treatment response of patients with bipolar I disorder. We hypothesized that anxiety, particularly with panic-like symptoms, would be associated with higher rates of nonremission and a longer time to remission.
We examined the first 124 patients to enter the protocol for Maintenance Therapies in Bipolar Disorder (NIMH grant MH-29618; Dr. Frank, primary investigator), an ongoing randomized trial testing the efficacy of interpersonal and social rhythm therapy (18) as an adjunctive treatment for bipolar I disorder.
In general, participants were required to 1) meet Research Diagnostic Criteria (19) for bipolar I disorder, 2) be in at least their third discrete affective episode, and 3) be between the ages of 21 and 65. In addition, patients had to meet severity criteria, defined as a score of 15 or higher on either the 17-item Hamilton Rating Scale for Depression (20) or Bech-Rafaelsen Mania Scale (21), and their most recent previous episode must have occurred less than 5 years prior to the index episode. Exclusionary criteria were 1) a lifetime diagnosis of schizophrenia, borderline personality disorder, or antisocial personality disorder; 2) chronic drug or alcohol abuse; 3) more than four affective episodes per year; 4) pregnancy; or 5) medical illness that would preclude protocol pharmacotherapy. Twenty-one patients entered the protocol through a separately randomized, expanded criteria cell that allowed the inclusion of patients who met criteria for schizoaffective disorder, manic type (N=7); had fewer than two previous affective episodes (N=5); reported no clear-cut period of remission (N=4); were taking a mood stabilizer other than lithium (N=3); met criteria for rapid cycling (N=1); or whose most recent episode occurred more than 5 years prior to the index episode (N=1). For the present analyses, patients with bipolar disorder and patients with schizoaffective disorder were combined because the course of their disorders was found to be very similar (22). All patients gave written informed consent after receiving a complete description of the study.
The first 64 patients were diagnosed with the Schedule for Affective Disorders for Schizophrenia (SADS) (23), and the remaining 60 patients were diagnosed with the Structured Clinical Interview for DSM-IV—Patient Edition (SCID-P) (24). Interviewers were trained bachelor’s- or master’s-level nurses and social workers experienced in the assessment of patients with bipolar disorder. Formal interrater reliability was not obtained, but every interview was reviewed for accuracy by the principal investigators and the program coordinator (E.F. and A.G.M). General functioning, as measured with the Global Assessment Scale (GAS) (25), age at onset of bipolar disorder, and number of episodes were also assessed by the diagnostic interviewer.
Prior to treatment, patients were administered the 25-item Hamilton depression scale (26), Bech-Rafaelsen Mania Scale (21), and Social Attainment Scale (27). Medication side effects were monitored by using the 22-item Somatic Symptoms Checklist developed for the present treatment study. (The measure is available from Dr. Frank.) The checklist, administered by the treating psychiatrist at each clinic visit, required the physician to rate the patient’s perception of the severity of 22 somatic side effects of medication on a 3-point scale (0=not present, 1=present but tolerable, or 2=present but causing significant distress or incapacity).
Patients were assessed by an independent evaluator at each clinic visit. Patients who achieved average scores of 7 or lower on the 17-item Hamilton depression scale and Bech-Rafaelsen Mania Scale for 4 consecutive weeks were classified as having achieved remission. Time to remission was defined as the number of weeks from the first treatment session to the end of this 4-week period of stabilization.
Twenty-seven patients were classified as nonremitters either because they did not achieve remission as defined above (N=11), were clearly noncompliant in taking their medication (N=5), or dropped out of the acute phase for reasons that were not obviously related to external circumstances (e.g., a move out of the Pittsburgh area) (N=11). Patients generally were withdrawn if they had not achieved remission after 24 weeks of acute treatment. However, in some cases, patients were permitted to continue the acute treatment following the guidelines explicated in the study protocol. Patients who continued beyond 24 weeks of acute treatment included those patients whose polarity had switched and those patients who were considered to have a reasonable prospect of remission. These decisions were based on continuous and careful discussions among the treatment team and continuous reassessments of patients’ symptoms. Five patients withdrew for external reasons and were therefore not classified as nonremitters.
Patients were randomly assigned to acute-phase treatment with either interpersonal and social rhythm therapy or intensive clinical management. Pharmacotherapy guidelines were constant across treatment conditions, and the pharmacotherapy algorithms applied in the present study have been detailed previously (28). Briefly, protocol pharmacotherapy began with lithium, but neuroleptics, antidepressants, anxiolytics, and other mood stabilizers could be used alternatively or adjunctively according to specified algorithms for mania and depression. Initial lithium doses ranged from 600 to 900 mg/day for depression and from 900 to 1200 mg/day for mania. Doses were increased as rapidly as clinically tolerated until target serum levels of 0.8 to 1.0 mmol/liter were achieved, although levels as high as 1.4 mmol/liter were permitted for the treatment of acute mania. Blood serum levels were measured 12 hours postdose at regular intervals throughout the acute treatment phase.
Potential Correlates of Outcome
Potential anxiety-related correlates of outcome included 1) history of panic attacks, 2) diagnosis of lifetime threshold or subthreshold anxiety disorder as per the SADS or SCID-P, and 3) baseline psychological and somatic anxiety as per the Hamilton depression scale. History of panic attacks was obtained from a review of diagnostic interviews. Both the SADS and SCID-P require interviewers to ask whether the patient has ever experienced a panic attack, and a positive response is followed by detailed questions about the symptoms experienced during the episode to ascertain that the criteria for a panic attack were met. The number of patients who met the lifetime threshold or subthreshold criteria for panic disorder was too small (seven remitters and six nonremitters) to allow meaningful analysis.
Based on a review of the literature, we selected a range of non-anxiety-related variables as other potential correlates of outcome. These included age, sex, marital status, education, employment status, current or past alcohol or substance use disorder, interepisode psychotic symptoms, age at onset and duration of the bipolar disorder, number of previous depressive and manic episodes, duration of index episode at study entry, and primary polarity treated during the acute phase. Because patients’ polarity may change during the course of treatment, and because we were interested in examining factors related to treatment outcome, we used the polarity that represented the primary focus of treatment during the acute phase, rather than the polarity at the moment of study entry, as a potential correlate of outcome. Baseline symptom correlates included depression (25-item Hamilton depression scale), mania (Bech-Rafaelsen Mania Scale), global functioning (GAS), and social and vocational functioning (Social Attainment Scale).
Potential correlates of remission (N=92) versus nonremission (N=27) were examined by using stepwise logistic regression analyses with backward deletion. Because the design of the present study specifies that patients who respond to the acute phase are treated until remission is achieved, time to remission is the main indicator of outcome in the acute treatment. Potential correlates of time to remission were identified by using stepwise Cox proportional hazards regression models with backward deletion. The latter analyses included the total intent-to-treat study group (N=124). Data from patients who withdrew prior to achieving remission (N=32) were censored at the time of withdrawal.
Bivariate correlation analyses revealed that two of the three anxiety variables were strongly associated with one another (r=0.36, N=124, p≤0.001 and r=0.25, N=124, p=0.006). In order to reduce overlap among the covariates included in our regression analyses, we combined the three anxiety variables into a single, dichotomous composite score reflecting past or present anxiety symptoms as shown by a 1) history of panic attacks, 2) lifetime threshold or subthreshold anxiety disorder, or 3) score of 3 or higher on either the psychological or somatic anxiety item of the Hamilton depression scale. Because we were particularly interested in the potential adverse effects of panic-related anxiety, we conducted two sets of regression analyses for each question to be addressed—one including history of panic attacks and the other including the anxiety composite variable as a covariate.
In order to further reduce the number of covariates, we first conducted a series of preliminary univariate tests comparing remitters and nonremitters on all of the potential correlates and examining the association between time to remission and all of the potential correlates. Only those variables that differentiated remitters from nonremitters and those that were associated with time to remission, with alpha set at ≤0.10 (two-tailed), continued to be considered for inclusion as covariates in the primary multivariate analyses.
Preliminary analyses were also conducted to check for potential differential effects of interpersonal and social rhythm therapy and clinical management on remission status and time to remission. The percentages of patients who were classified as nonremitters were very similar for interpersonal and social rhythm therapy and clinical management (χ2=0.39, df=1, p=0.53), and a survival analysis comparing time to remission for patients treated with interpersonal and social rhythm therapy versus clinical management yielded nonsignificant results (log rank=0.09, df=1, p=0.76). Consequently, treatment group was not statistically controlled for in subsequent analyses.
Primary polarity treated during the acute phase was coded with two dummy variables to indicate depression or mixed polarity. Mania was treated as the reference polarity. Where indicated, scores were logarithmically transformed so as to enhance the fit between their distributions and the assumptions of the analyses performed.
Fifty-six patients of the total study group (45.2%) experienced present or past anxiety symptoms as assessed with the composite variable. Of these, 34 patients had a history of panic attacks, 28 met criteria for at least one lifetime subthreshold or threshold anxiety disorder (range=1–3: panic disorder [N=13], posttraumatic stress disorder [N=6], specific phobia [N=5], OCD [N=4], generalized anxiety disorder [N=4], social phobia [N=3], and substance-induced anxiety disorder [N=1]), and 21 scored 3 or higher on either the psychological or somatic anxiety item of the Hamilton depression scale. Of the 56 patients with anxiety, the majority were treated for depressive (N=35 versus 31 nonanxious depressed patients) or mixed (N=14 versus 13 nonanxious mixed patients) symptoms rather than manic (N=7 versus 24 nonanxious manic patients) symptoms. Similarly, of the 34 patients with a history of panic attacks (27.4% of the total study group), the majority were treated for depressive (N=23 versus 43 nonanxious depressed patients) or mixed (N=7 versus 20 nonanxious mixed patients) symptoms rather than manic (N=4 versus 27 nonanxious manic patients) symptoms.
Correlates of Nonremission
Preliminary t tests and chi-square tests revealed that nonremitters were more likely to report a history of panic attacks (χ2=6.13, df=1, p=0.01), current or past anxiety as per the composite variable (χ2=6.53, df=1, p=0.01), more severe depression (25-item Hamilton depression scale: t=4.43, df=48, p<0.001), and a greater number of previous depressive (t=3.83, df=117, p<0.001) and manic (t=2.83, df=36, p=0.008) episodes, and tended to report more severe symptoms of mania (Bech-Rafaelsen Mania Scale: t=1.98, df=52, p=0.053). In addition, they were more likely to be treated for depression during the acute phase (χ2=9.87, df=2, p=0.007).
Examination of the potential covariates to be entered into the logistic regression model showed that being treated primarily for depression (versus being treated primarily for mania) was strongly associated with baseline scores on the 25-item Hamilton depression scale (r=0.61, N=119, p<0.001) and the Bech-Rafaelsen Mania Scale (r=–0.71, N=119, p<0.001). To reduce overlap among the covariates, Hamilton depression scale and Bech-Rafaelsen Mania Scale scores were dropped from the analyses. Multicollinearity was acceptable after bivariate correlations were controlled.
With history of panic attacks included as a potential correlate of nonremission, only greater number of depressive episodes and history of panic attacks emerged as significant covariates in the final regression model (t1). With anxiety composite scores included as a potential correlate and history of panic attacks omitted, only a treatment focus of depression versus mania and number of manic episodes emerged as significant covariates in the final model (Wald χ2=4.64, df=1, p=0.03 and Wald χ2=4.04, df=1, p=0.04, respectively).
Correlates of Time to Remission
Preliminary correlation coefficients and univariate analysis of variance showed the following variables to be associated with weeks to remission with alpha set at ≤0.10: history of panic attacks (r=0.28, N=92, p=0.008), present or past anxiety symptoms as per the composite variable (r=0.33, N=92, p=0.001), 25-item Hamilton depression scale scores (r=0.33, N=92, p=0.001), Bech-Rafaelsen Mania Scale scores (r=–0.22, N=92, p=0.04), duration of index episode (r=0.29, N=92, p=0.005), lifetime alcohol or substance use disorder (r=0.25, N=92, p=0.02), and polarity treated during the acute phase (F=19.61, df=2, 92, p<0.001).
Because examination of covariates to be included in the Cox regression model revealed a strong overlap between primary polarity treated and scores on the 25-item Hamilton depression scale and the Bech-Rafaelsen Mania Scale, these two variables were excluded from the multivariate analyses. None of the remaining variables exhibited significant deviance from the proportional hazards assumption.
With history of panic attacks included as a potential correlate of time to remission, depressive and mixed versus manic symptoms emerged as significant covariates of longer time to remission in the final model, although history of panic attacks also tended to be associated with a longer time to remission (t1) (overall χ2=49.47, df=3, p<0.001). With anxiety composite scores included as a potential correlate of time to remission, treatment of depressive and mixed versus manic symptoms and current or past anxiety as per the composite variable proved to be associated with longer time to remission (t1) (overall χ2=55.04, df=3, p<0.001).
To determine the potential differential impact of anxiety in relation to patients’ affective state during acute treatment, we conducted separate post hoc Kaplan-Meier survival analyses testing for the effects of anxiety in patients with manic, depressed, and mixed symptoms. Current or past anxiety was associated with significantly longer time to remission only in patients with mania (F1), whereas history of panic attacks was associated with a significantly longer time to remission only in patients with depression (F2).
Exploratory analyses were conducted to examine potential mechanisms by which anxiety may impede outcome. One possibility is that anxiety decreases patients’ tolerance for medication side effects, thereby causing them to be more likely to receive subtherapeutic medication doses (either because they are less compliant or are prescribed lower doses) or causing them to require more frequent medication changes to reduce potentially anxiety-evoking side effects.
Consequently, we first compared anxious and nonanxious patients on the severity of medication side effects. Scores on the Somatic Symptoms Checklist were averaged across all assessment points during the acute phase. As predicted, patients with current or past anxiety reported more severe side effects (mean=6.72, SD=3.02) than those patients without anxiety (mean=4.89, SD=2.88) (t=3.64, df=122, p<0.001). Similarly, patients with a history of panic attacks reported more severe side effects (mean=6.80, SD=3.01) than those patients without panic attacks (mean=5.30, SD=3.01) (t=3.05, df=78, p<0.003).
Medication Compliance or Tolerance
Compliance or tolerance to medication was assessed by examining lithium blood serum levels and prescribed lithium doses. To explore whether anxious patients had lower lithium blood serum levels, we first compared the two groups on average blood levels of lithium across the acute phase. Second, because lower lithium levels may be a result of either lack of compliance or low prescribed doses, we examined potential differences in compliance, as assessed by the stability of blood serum level/dose ratios (blood serum levels divided by doses of lithium) (29), as well as prescribed lithium doses averaged across the acute phase.
Patients with a history of panic attacks showed lower levels of blood lithium (mean=0.77, SD=0.11) than patients without panic attacks (mean=0.83, SD=0.14) (t=1.94, df=111, p=0.05). However, the lithium levels of patients with current or past anxiety (mean=0.80, SD=0.14) did not differ from the levels of patients without anxiety (mean=0.83, SD=0.13) (t=1.25, df=111, p=0.21).
To assess medication compliance, we calculated the standard deviation of the blood serum level/dose ratios for each patient and compared the average standard deviations of anxious and nonanxious patients. We expected that anxious patients would show a greater variability of blood serum level/dose ratios, as shown by larger standard deviations. This did not prove to be the case. The standard deviations of blood serum level/dose ratios of patients with current or past anxiety (mean=1.43, SD=0.67) did not differ from the standard deviations of patients without anxiety (mean=1.26, SD=0.51) (t=1.48, df=106, p=0.14). Nor did the standard deviations from blood serum level/dose ratios of patients with a history of panic attacks (mean=1.34, SD=0.59) differ from the standard deviations of patients without panic attacks (mean=1.33, SD=0.59) (t=0.01, df=106, p=0.99).
Finally, prescribed lithium doses of patients with current or past anxiety (mean=1155 mg, SD=329) did not differ from the doses of patients without anxiety (mean=1246 mg, SD=322) (t=1.55, df=117, p=0.12). Nor did prescribed lithium doses differ between patients with panic attacks (mean=1172 mg, SD=344) and patients without panic attacks (mean=1219 mg, SD=321) (t=0.83, df=117, p=0.41).
Number of Medications Prescribed
We hypothesized that anxious patients would require a larger number of different medications to achieve remission, perhaps to avoid intolerable side effects, and/or more aggressive pharmacotherapy because of their lack of improvement. Medications were categorized as mood stabilizers, neuroleptics, antidepressants, and anxiolytics.
Compared to patients without current or past anxiety, patients with anxiety required a larger total number of different medications, either sequentially or in combination (mean=3.91, SD=2.64 versus mean=2.83, SD=1.80) (t=2.12, df=55.1, p=0.04). This occurred, in part, because anxious patients required a larger number of antidepressant trials (mean=1.32, SD=1.75 versus mean=0.69, SD=0.99) (t=2.16, df=90, p=0.03). The two groups did not differ in the number of different mood stabilizers, neuroleptics, or anxiolytics needed to achieve remission.
Compared to patients without a history of panic attacks, patients with panic attacks required a larger total number of different medications (mean=4.26, SD=2.88, versus mean=2.96, SD=1.92) (t=2.15, df=90, p=0.04), either sequentially or in combination. No differences emerged in the number of mood stabilizers, neuroleptics, antidepressants, or anxiolytics required to achieve remission.
We examined a range of demographic, psychiatric history, and symptom variables previously shown to be related to treatment outcome and found anxiety to be a clinically meaningful correlate of poor response to the acute treatment of bipolar disorder. First, a history of panic attacks, together with a greater number of previous depressive episodes, was significantly associated with nonremission. Second, current or past anxiety symptoms as assessed with a composite variable emerged as a significant correlate of longer time to remission, although the effects of history of panic attacks on time to remission did not reach statistical significance. Inspection of raw data indicated that anxious patients required longer time to remission than nonanxious patients, regardless of their primary affective state during acute treatment. However, post hoc survival analyses showed that anxiety as per the composite variable was associated with a significantly longer time to remission only in patients treated for mania, whereas a history of panic attacks emerged as a significant correlate of longer time to remission only in patients treated for depression. Nonsignificant results yielded by these post hoc analyses may, in part, be attributed to reduced statistical power due to limitations of the group size.
Taken together, our findings are consistent with studies demonstrating a poorer response to treatment in patients with unipolar depression and anxiety (2, 8–12). They also fit previous observations indicating that a significant percentage of individuals with bipolar disorder experience concomitant anxiety (13–15), which is associated with greater clinical severity (15).
Because anxiety was not a focus of our research interest at the inception of the present treatment trial, our assessment was limited to measures that imperfectly capture its multifaceted symptoms. Most notably, interrater reliability for the diagnosis of anxiety disorders and history of panic attacks was not obtained. However, given that our diagnostic interviewers had considerable expertise in the assessment of affective as opposed to anxiety disorders and given that the main purpose of the diagnostic interviews was to establish eligibility criteria for participation in the present treatment trial—none of which was related to anxiety—it seems more likely that the presence of anxiety disorders was underrecognized rather than overdiagnosed. A more thorough assessment of anxiety might yield even stronger evidence for the impact of anxiety on treatment response in bipolar disorder and would allow us to determine which aspects of anxiety (e.g., panic-related fear of somatic arousal or worry) are related to treatment response.
Exploratory analyses were conducted to examine the mechanisms by which anxiety impedes treatment outcome. Our own clinical observations suggest that anxiety decreases patients’ tolerance for medication side effects, thereby causing them to be less compliant, be prescribed lower doses, and/or require more frequent medication changes to reduce potentially anxiety-evoking side effects—all of which are factors that are likely to impede improvement.
We found mixed support for this hypothesis. Patients with current or past anxiety as per the composite variable and patients with a history of panic attacks reported significantly more severe medication side effects, averaged across the entire acute treatment phase. Furthermore, although patients with a history of panic attacks were not prescribed lower lithium doses, they showed significantly lower lithium blood serum levels. However, blood serum levels of patients with current or past anxiety did not differ from levels of nonanxious patients. Finally, patients with current or past anxiety and patients with a history of panic attacks required a significantly larger number of medications in order to achieve remission. The latter finding fits our hypothesis that anxious patients prompted more medication changes in an attempt to reduce side effects. Alternatively, anxious patients’ larger number of medications may reflect their lack of improvement or some other factor, such as physiological intolerance (30). We attempted to address this question but were unable to reliably determine the reason for patients’ medication changes. Contrary to our expectation, anxious patients were not less compliant to lithium nor were they prescribed lower lithium doses. It should be noted that blood serum levels of lithium were obtained for 50.8% of all clinic visits, not allowing us to rule out some undetected bias in the assessment of lithium compliance. Given the exploratory nature of our analyses aimed at understanding the potential mechanisms by which anxiety may impede outcome, a Bonferroni correction for the number of tests conducted was not applied. These findings should be interpreted with caution.
Clearly, the present results do not allow us to reject alternative explanations for the apparent adverse effects of anxiety. Rather than playing a causal role in hampering bipolar patients’ response to treatment, anxiety may simply be a marker of other risk factors or, more generally, a nonspecific severity marker. Furthermore, we did not take into account all of the potential correlates of outcome. However, of the variables that have been shown or may be expected to be associated with poor outcome (e.g., low socioeconomic status or personality disorders), few can be as readily addressed as anxiety. If the adverse effects of anxiety are confirmed in independent studies, strategies that enhance the tolerance of medication side effects for anxious patients with bipolar disorder may have the potential to enhance treatment efficacy. Such strategies might include a substantial reduction of initial medication doses, much slower than normal upward titration schedule, and symptomatic treatment of anxiety-evoking side effects. If psychotherapy is offered in addition to medication, an effective intervention might include one of the anxiety-specific psychotherapies, in addition to elements aimed at mood symptoms. Thus, although clinicians cannot control patients’ polarity or the number of previous affective episodes, there are numerous ways in which they could address the potentially negative impact of anxiety on the treatment outcome of bipolar disorder.
Received May 21, 1999; revision received Nov. 19, 1999; accepted Nov. 29, 1999. From the Western Psychiatric Institute and Clinic, Department of Psychiatry, University of Pittsburgh School of Medicine. Reprints are not available. Address correspondence to Dr. Frank, Western Psychiatric Institute and Clinic, 3811 O’Hara St., Pittsburgh, PA 15213; firstname.lastname@example.org (e-mail). Supported by NIMH grants MH-29618, MH-30915, MH-16804, and MH-01675; and the Stanley Foundation for the Innovative Treatment of Bipolar Disorder.
Time to Remission of Bipolar Disorder for Patients With and Without Current or Past Anxiety Symptoms, Based on Kaplan-Meier Survival Analysis
aLog rank=4.37, df=1, p=0.04.
bLog rank=2.95, df=1, p=0.09.
cLog rank=1.45, df=1, p=0.29.
Time to Remission of Bipolar Disorder for Patients With and Without History of Panic Attacks, Based on Kaplan-Meier Survival Analysis
aLog rank=0.19, df=1, p=0.66.
bLog rank=4.32, df=1, p=0.04.
cLog rank=0.85, df=1, p=0.36.